InitRunner

You've likely built agents before, so you understand the process: setting up a framework, creating a model interface, defining tools with code, adding memory and RAG, and establishing logging, repeating these steps for each new agent.

InitRunner condenses this into a YAML file.

By installing the package, you can define your agent's model, system prompt, tools, and guardrails in a role.yaml file, and then run it directly from your terminal. This allows you to configure agents in YAML instead of writing the code to wire them together yourself.

A single role file can be executed in four ways: as a one-time command, an interactive REPL, an autonomous agent that can plan and execute multi-step tasks, or as an OpenAI-compatible API server.

RAG, long-term memory, and token budget management are all specified within the same YAML file, rather than being assembled within the application code. It supports several providers (OpenAI, Anthropic, Google, Groq, Mistral, Cohere, Bedrock, xAI, Ollama) right out of the box, and switching between them requires only a single line of code.

Agents come equipped with tools for common tasks, eliminating the need to build them yourself: Git operations, filesystem access, HTTP requests, web scraping, shell commands, SQL queries, Python execution, email, web search, Slack messaging, scheduling, and MCP server integration. Custom tools can be added as a single file. Each interaction is recorded in an append-only audit log, and a web dashboard and terminal UI allow you to monitor runs without having to sift through stdout.

For scenarios where one agent isn't sufficient, a compose.yaml file can wire multiple agents together into a single workflow, with routing, shared context, and structured output between them.